datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
phonemizer-dicts
Phonemizer Dicts
Pre-generated IPA dictionaries for GPL-free text-to-phonemes lookup.
Files
en-us.tsv — 124K English (US) words, tab-separated word<TAB>IPA
Provenance
Generated by running espeak-ng over an English wordlist. The TSV is program output; espeak-ng source (GPL-3.0) is not redistributed here.
Regeneration
See scripts/generate-espeak-dict.py in the tts-rd-team repo.
Phonemized-UD
Phoneme-UD: A Multilingual Phonemized Universal Dependencies Corpus for 34+ Languages
G2P+ Phonemizer
We use G2P+ to phonemize Universal Dependencies. Here is an example usage:
# Install required packages
!apt-get install -y espeak-ng
!pip install phonemizer g2p-plus
# Set the environment variable from Python
import os
os.environ["PHONEMIZER_ESPEAK_LIBRARY"] = "/usr/lib/x86_64-linux-gnu/libespeak-ng.so.1"
# Now run your transcription
from g2p_plus import… See the full description on the dataset page: https://huggingface.co/datasets/suchirsalhan/Phonemized-UD.ghana-speech-phonemized
Ghana Speech (Phonemized)
Preprocessed multilingual speech dataset from
ghananlpcommunity/ghana-speech,
covering 42 Ghanaian and West African languages with IPA phoneme transcriptions.
This dataset contains only the metadata and phonemes — no audio.
Use it together with the source dataset to load audio on-the-fly.
Contents
Each language is a separate config (Parquet). Columns:
Column
Type
Description
id
string
Unique clip ID (matches audio IDs in the… See the full description on the dataset page: https://huggingface.co/datasets/ghananlpcommunity/ghana-speech-phonemized.eng_spa_owt_phonemized_fullghana-speech-lfn-phonemizedtarteel-ai-EA-DI-phonemized-Finalcatalan-dataset-phonemizedeng_spa_owt_phonemized_full_romanizedvoxpopuli-spanish-espeak-phonemizedcatalan-dataset-phonemized-splitxlsum_unseen_phonemized_romanizedopentts-phonemized-sentencesPhonemized version of https://huggingface.co/datasets/speech-uk/text-to-speech-sentences with some additional fields.
tarteel-ai-EA-DI-phonemized10M_phonemized_German_dataset
Small Phonemized German dataset
Description
This dataset is a smaller and phonemized version of Bastian Bunzeck's dataset (Bunzeck et al., 2025) and contains approximately 10 million words.
The phonemization was done using Phonemizer by Bernard & Titeux, 2021.
This dataset was used to train the Phoneme-based German babyLlama model.
Source Composition
The dataset has been compiled from the following sources:
Source
Description
Words… See the full description on the dataset page: https://huggingface.co/datasets/CLAUSE-Bielefeld/10M_phonemized_German_dataset.pl-bert-phonemized-pltparlertts_pony_speech_phonemizedcloning_pairs_text_non_phonemizedphonemized_transcriptions